[nm-wg] Does request/response support "Push"?

["Jeff W. Boote" <boote@internet2.edu>]

I'm currently working on expanding our Pipes software so that each
administrative domain can set up regular tests within its own domain. This will
allow administrative domains to choose a subset of measurement hosts to do tests
with hosts in other administrative domains. (This would also be used to allow
meta-domains such as the paths of interest for a given application community.)
In these cases, I would typically expect both domains to want to see the results
of these measurements.

In working on this, it occurs to me that there is a piece the current
request/response "model" doesn't really address. I am not yet sure if I think it
should be addressed, but I'm curious what others think about it:

Basically, I have setup our software so that most things are event driven. In
other words, the measurement host sends data to a database when it has data to
send. The database does not poll the measurement host for data. I didn't think
too much of this in the past, but now that I'm thinking about multiple domains
being interested in the results of a measurement, I am planning on having the
measurement host send data to multiple databases when the data becomes
available. This type of a data exchange is not directly addressed by the current
request/response model. In the past I have always thought of this kind of data
exchange as an internal issue and didn't really think it would be too important
to have a defined schema for it. However, in the world of multiple measurement
domains trying to share results...

To support this type of model, I think it might be advantageous to have a schema
for a data push. I think it would be fairly similar to the "response" schema
(perhaps nearly identical - I need to think more on this). In some ways, I guess
you could think of this as a response to "all future measurements" perhaps
matching some given criteria. However, in terms of being able to configure up
new measurement points without having to know about all of them in advance, I
really prefer more of a "push" model than a "pull" model. The database would of
course need to have some kind of authentication and criteria for deciding what
measurements it is going to accept and from whom...

Anyone else think this might be worth doing? If others think it is worth doing,
I would think it can probably wait until the current schema's are more mature.
But it might be worth giving at least a little thought in the response schema on
what is/is not applicable to a "Data Push". The more recent developments to make
the current schemas more "modular" definitely helps in this regard.

(If no one wants to deal with this right now (or even ever) because there is
plenty on the wg plate, I totally understand. I just bring it up because I'm
having to deal with some of these issues right now, and I'm curious what others
in this forum might think.)

jeff